AMD Details New Model Numbering System For 2023 Mobile Processors

[Shtirlic]

Why not to put year in the front like: 23640U, 22735U and get rid of the stupid 7,8,9 ?

[Luke_Wolf]

The old scheme, without using additional digit for "architecture", let people deduce the architecture from the first digit. (Not totally consistent but usually okay.) The marketing people may like new products to always contain bigger number, but sane buyers will wish old model refreshes be named as proper old model refreshes. Whatever they refreshed or improved outside core architecture should have been minor details that only deserve the "feature isolation" digit at the end.

The old scheme never had architecture as part of it. the first number was always the marketing model year and this is consistent across all vendors, they play silly games with changing up the marketing model year but that's always been the case.

Relative Performance within a Marketing Model Year has basically always been the second relevant number again across all vendors

A further Performance Segementation Number has sometimes but not always been done, again cross vendor

AMD is introducing for the first time of any of the big 3 (AMD, Intel, Nvidia) an Architecture Number which is something they should be lauded for not bitched at about because previously this information was hidden from consumers who didn't do their research.

The first two digits are literally staying the same as how things have been done, the only things that are changing are reintroducing the performance segmentation number, and ADDING a number to identify what parts are using what generation of hardware. What really is so hard to comprehend about this?

[arQon]

For me this new numbering looks completely f***ed.

Agreed. Especially since the whole system falls apart in 3 years. But they're made up by marketing people, so you can't really expect much. :P

> and why market segment number is more important than the architecture?

Because it is? 8 of last gen's cores still easily outperforms 6 of this gen's for everything except single-threaded code, for every CPU since the last time AMD was competitive. Zen4 is potentially the *only* exception to that, and even so, a Zen3 59xx will still easily outperform a 6-core Zen4 for any task that someone buying a 32+T chip cares about.

[binarybanana]

schmidtbag
What's wrong with buying one of those 65W models and limiting it to 35W or whatever? They support ECC (given the right mainboard) and have enough IO for smaller servers. Makes no sense to release a separate SKU with gimped specs when yields are good and price already low.

[Rabiator]

There is a big market for such servers but as you alluded to, there's a hell of a lot of competition, and not just from Intel. AMD is raking in cash being basically the only sensible option for those 280W monster chips. From what I understand, they're still struggling to keep up with production in the markets they already target, so it makes sense they don't make a Epyc 5000 series, at least for now.

(Edit):
Talking about small servers, AMD already has the Pro 5650G and the Pro 5750G. That is the up-to-date equivalent to entry-level Xeons.
With the market demand being somewhat diminished these days, quantity should stop being an issue soon.

[rmfx]

That is atrocious.

MY nomenclature:
A (26 gen to go no looking similar to each other)
7 ( 3/5/7/9 for the tier)
8 ( the model variant)
0 ( 0 for the first release, 5 for the optional refresh)
m ( optional letters for mobile and co)

…with this, you have 50 years to go with consistency instead of all the useless marketing morons screwing us.

[schmidtbag]

Talking about small servers, AMD already has the Pro 5650G and the Pro 5750G. That is the up-to-date equivalent to entry-level Xeons.
With the market demand being somewhat diminished these days, quantity should stop being an issue soon.

Those are workstation CPUs, not small-server CPUs. As a CPU, there's hardly any difference, but the platforms you'll find them in are quite different.

What's wrong with buying one of those 65W models and limiting it to 35W or whatever? They support ECC (given the right mainboard) and have enough IO for smaller servers. Makes no sense to release a separate SKU with gimped specs when yields are good and price already low.

Last I checked, AMD isn't selling any new 65W Epycs. As I mentioned to Rabiator, we're talking about servers, not workstations. Stuff that has far higher production numbers and likely to be rack-mounted.

[torsionbar28]

schmidtbag
What's wrong with buying one of those 65W models and limiting it to 35W or whatever? They support ECC (given the right mainboard) and have enough IO for smaller servers. Makes no sense to release a separate SKU with gimped specs when yields are good and price already low.

False, the Ryzen ECC support is crippled and mostly non existent. It's partially enabled on the PRO variant only, which is an OEM exclusive, not available at retail. And even on the PRO variant, the ECC feature is supposedly "enabled", but you're forced to take it on a matter of faith, as the ECC reporting, arguably the most important part of it all, does not exist on Ryzen. Not to mention the spotty mobo/bios support.

In summary, IF you can find a pro variant (ahem, ebay), and you can locate one of the very few mobo's that advertises ECC support, and you find some compatible ECC sticks (many mobo's that list ECC as supported don't officially qualify any ECC sticks so you're left to guess), Even after all of that, you still don't have ECC reporting! So if you have a bad stick that's constantly correcting single bit errors - you have no knowledge of it. And if you have a really bad stick that is experiencing uncorrectable multi-bit errors - you have no knowledge of it. Your crashing programs and corrupt data are your only indicators, which is hardly any better than a non-ECC machine.

So no, Ryzen is not suitable at all - not even Ryzen PRO - for an edge server use case. Xeon E-2300 is the only option, until AMD has the capacity to release a low TDP EPYC 5000 series.

[deusexmachina]

False, the Ryzen ECC support is crippled and mostly non existent. It's partially enabled on the PRO variant only, which is an OEM exclusive, not available at retail. And even on the PRO variant, the ECC feature is supposedly "enabled", but you're forced to take it on a matter of faith, as the ECC reporting, arguably the most important part of it all, does not exist on Ryzen. Not to mention the spotty mobo/bios support.

In summary, IF you can find a pro variant (ahem, ebay), and you can locate one of the very few mobo's that advertises ECC support, and you find some compatible ECC sticks (many mobo's that list ECC as supported don't officially qualify any ECC sticks so you're left to guess), Even after all of that, you still don't have ECC reporting! So if you have a bad stick that's constantly correcting single bit errors - you have no knowledge of it. And if you have a really bad stick that is experiencing uncorrectable multi-bit errors - you have no knowledge of it. Your crashing programs and corrupt data are your only indicators, which is hardly any better than a non-ECC machine.

So no, Ryzen is not suitable at all - not even Ryzen PRO - for an edge server use case. Xeon E-2300 is the only option, until AMD has the capacity to release a low TDP EPYC 5000 series.

This is really terrible. I wonder how the ECC support is on POWER9 and POWER10s.

[chithanh]

In summary, IF you can find a pro variant (ahem, ebay), and you can locate one of the very few mobo's that advertises ECC support, and you find some compatible ECC sticks (many mobo's that list ECC as supported don't officially qualify any ECC sticks so you're left to guess), Even after all of that, you still don't have ECC reporting! So if you have a bad stick that's constantly correcting single bit errors - you have no knowledge of it.

I think that is incorrect. Only for APUs, Pro is required for ECC support. Non-Pro CPUs support ECC too.
Also lots of AM4 mobos advertise ECC support in the specs. ASRock and Biostar all, ASUS only higher end, Gigabyte all since 500 series (previously only higher end), MSI none.

Reporting also works via amd64_edac driver. There was one ASRock mobo where for some releases a BIOS bug prevented it, but to my knowledge it is fixed now. There was an article by Hardware Canucks a while back where they verified it (ignore the part about halting the system on uncorrectable errors, the author was misinformed on default Linux behavior): https://hardwarecanucks.com/cpu-moth...zen-deep-dive/

If you have a concrete example/report of an AM4 mobo where ECC is advertised but not working properly, please provide a link.